The adversarial joint source-channel problem

Yuval Kochman, Arya Mazumdar, Yury Polyanskiy

Abstract: This paper introduces the problem of joint source-channel coding in the setup where channel errors are adversarial and the distortion is worst case. Unlike the situation in the case of stochastic source-channel model, the separation principle does not hold in adversarial setup. This surprising observation demonstrates that designing good distortion-correcting codes cannot be done by serially concatenating good covering codes with good error-correcting codes. The problem of the joint code design is addressed and some initial results are offered.